Streamflow Simulation Based on Pre-Processing-Based Hybrid Model: SSA-AR Model

Abstract An accurate medium-term streamflow forecast is one of the significant functions for managing and planning water resources. Considering the characteristics of trend, periodicity, and stochasticity of streamflow into account. So, this research aims to develop a new strategy, including a singular spectrum analysis (SSA) technique and a linear autoregressive (AR) model to predict monthly Tigris River streamflow data with three scenarios. The first scenario applies the SSA to decompose the normalised and cleaned time series into different signals (i.e., trend, seasonal, stochastic, and noise), then reconstruct the signals without noise and use the AR model to simulate the new time series. The second scenario employs the AR model to forecast each signal of streamflow without noise separately. The simulated time series was obtained by summing each predicted signal. The third scenario uses the AR model to simulate the raw data. Based on several statistical tests, the comparative analysis reveals that the first and second scenarios were much more accurate than the third ones. The second scenario is the best, reaching RMSE = 0.1223 (m 3 /s) and MAE = 0.0913 (m 3 /s) in the testing phase. The novelty of this study lies in the comparative evaluation of three SSA-AR modelling scenarios and the finding that forecasting decomposed components individually leads to superior accuracy compared to conventional approaches. The results are of substantial significance to the Ministry of Water Resources in managing and planning freshwater resources amid growing water demand.